IQN — Implicit Quantile Networks.
hardAnswer
- Instead of predicting discrete atoms (C51) or fixed quantiles (QR-DQN), IQN learns to output ANY quantile τ ∈ [0, 1] given as input to the network.
- Input concatenates state + embedding of τ.
- Enables sampling arbitrary many quantiles at inference for risk-sensitive control.
- Cleaner formulation than QR-DQN's fixed quantiles.
- State-of-the-art distributional RL circa 2018-2020.
Check yourself — multiple choice
- Same as C51
- IQN: network outputs Q at ANY quantile τ ∈ [0, 1] via τ-embedding input; sample arbitrary quantiles at inference; cleaner than QR-DQN's fixed quantiles
- Random
- Not real
IQN: τ-embedded network outputs any quantile; flexible distributional RL.
#value-methods#deep-rl
Practise Reinforcement Learning
214 interview questions in this topic.